Upload
luzbel-ayala
View
231
Download
0
Embed Size (px)
Citation preview
7/27/2019 Lesvos Bakker Et Al
1/15
Agriculture, Ecosystems and Environment 105 (2005) 467481
www.elsevier.com/locate/agee
Soil erosion as a driver of land-use change
Martha M. Bakkera,*, Gerard Govers
b, Costas Kosmas
c, Veerle Vanacker
a,
Kristof van Oostb, Mark RounsevellaaDepartement de Geographie, Universite Catholique de Louvain, Place Pasteur 3, B-1348 Louvain-la-Neuve, Belgium
bLaboratory for Experimental Geomorphology, K.U. Leuven, Redingenstraat 16, B-3000 Leuven, BelgiumcAgriculturalUniversity of Athens, Iera Odos 75, Votanikos, GR-11855 Athens, Greece
Received 7 October 2003; received in revised form 13 July 2004; accepted 14 July 2004
Abstract
Although much research has been carried out on the crop productivity response to soil erosion, little is known about the role
of soil erosion as a driver of land-use change. Given, however, the some-times large erosion-induced reductions in crop yields, it
appears likely that erosion has a strong impact on land-use. Abandonment of arable land due to declining productivity is a land-
use change that may result from soil erosion. To test this hypothesis, the western part of Lesvos, Greece, was chosen as a case
study area. Lesvos has experienced accelerated erosion on marginal soils over the last century during which important land-use
changes have taken place. Of the 3211 ha that were under cereals in 1886, 53% (1711 ha) was converted to rangeland (only used
for extensive grazing) by the mid-20th century. At the same time, however, cereals partly returned to neighbouring areas thatwere previously rangeland, implying that certain processes at the local scale resulted in land becoming unsuitable in one place
and (relatively) more suitable in other places. In order to identify the relationship between these land-use changes and the
occurrence of soil erosion, erosion was modelled backwards for the period 18861996 and soil depths reconstructed for the time
when the land-use was assumed to have changed (the mid-1950s). A logistic regression was performed with soil depth, erosion
and slope as explanatory variables and land-use change as the response variable. Abandonment/reallocation of cereals was found
to be fairly well predicted by slope and soil depth. Path analysis showed erosion to be an important driver for the abandonment
and reallocation of cereals, although next to slope and soil depth it has little additional predictive value. Based on the logistic
model, it is anticipated that cereal cultivation in western Lesvos will probably be abandoned in the near future.
#2004 Elsevier B.V. All rights reserved.
Keywords: Soil erosion; Crop productivity; Land abandonment; Land-use change; Logistic regression; Path analysis; Lesvos
1. Introduction
* Corresponding author. Tel.: +32 10 47 85 06;
fax: +32 10 47 28 77.
E-mail address: [email protected] (M.M. Bakker).
Many studies have reported on the detrimental
effects of soil erosion on agricultural productivity.
A review of data collected in intense, mechanised
agricultural systems suggests that erosion reduces
0167-8809/$see front matter# 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.agee.2004.07.009
http://www.elsevier.com/locate/ageemailto:[email protected]:[email protected]://www.elsevier.com/locate/agee7/27/2019 Lesvos Bakker Et Al
2/15
468 M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481
productivity on average by about 4% for each 10 cm of
soil lost. Even higher reductions in productivity are
possible although this depends on the contrast between
top and subsoil properties (e.g. texture, fertility), andthe stage of the erosion process. For soils with rooting
depth restrictions (e.g. bedrock), productivity reduc-
tions become increasingly worse with incremental soil
depth reductions (Bakker et al., 2004).
Although the negative effect of erosion on crop
productivity has been extensively researched, little
research has been carried out on the effect of erosion-
induced reductions in crop productivity on land-use
change (Marathianou et al., 2000; Van Rompaey et al.,
2002). It may be hypothesised, however, that certain
types of land-use change, such as the replacement of a
land-use that puts high demands on soil by one with
lower demands, are caused by soil erosion. As soil
erosion is a process that varies strongly in space, one
may then anticipate a migration of land-use types with
high soil demands from strongly eroding sites to less
eroding sites, if the latter are available.
In Lesvos (Greece), an island suffering from high
erosion rates, important land-use change has occurred
during the past century. Of the 4582 ha that were under
cereals in 1886, 46% have been abandoned and are
currently used for extensive grazing. In the western
part of the island, where soils are shallower and theclimate is more arid, this situation is even more
pronounced. Here, a total of 3211 ha was under cereals
in 1886, of which 53% (1711 ha) was abandoned
(converted to rangeland) and 33% (1077 ha) changed
into other land-uses. While cereal production was
abandoned over large areas, other areas saw the
introduction of cereal cultivation, bringing the total
area under cereals in western Lesvos in 1996 back to
1822 ha, slightly more than half of the area in 1886.
The new cereal areas were previously under rangeland
(77%) and oak (Quercus spp.) plantations (16%). Only13% (418 ha) of the area that was under cereals in
1886 is currently still under cereals (see maps in
Marathianou et al., 2000).
Thus, besides a widespread abandonment of land
under cereals, rangelands were also replaced by
cereals at some locations. This suggests that exogen-
ous economic factors were not the (primary) cause of
cereal abandonment. If the cultivation of cereals were
no longer economically viable, production would have
been completely abandoned rather than being intro-
duced in some areas. It seems reasonable to conclude
that certain local scale processes were responsible for
the abandonment of some areas of cereal cultivation
making other areas relatively better suited for cereals,thereby causing a spatial shift in the location of
production.
A socio-economic explanation may exist for the
abandonment of cereals in the western hilly part of the
island. In the second half of the 20th century,
mechanised equipment was introduced and the former
agricultural systems based on human and animal power
were no longer able to compete (Kosmas et al., 1999).
However, some sources claim that the abandonment of
cereals was directly related to land degradation arising
from soil erosion. Referring to Lesvos,Kosmas et al.
(2000b) wrote: as soil was eroded, land use usually
shifted from agriculture to pasture due to increasingly
poor yields from the various agricultural crops, and
Marathianou et al. (2000) stated that land cultivated
with cereals was greatly reduced in eroded, hilly areas
where productivity was low.
When trying to assess the relationship between
erosion and land-use change, a problem that is often
encountered is that in general terms past agricultural
productivity has increased rather than decreased
because of the effects of technological innovations
and improved land management. However, theanalysis presented in this paper focuses on the spatial
patterns of land-use change, rather than the temporal
evolution of agricultural productivity. An investiga-
tion is made of the relationship between the spatial
pattern of abandonment and reallocation and the
severity of erosion during the period preceding a
change in land-use.
A further problem is confounding between the
effects of soil erosion and the direct effects of slope on
land-use. In this case confounding is the result of a
causal relationship between slope, erosion and soildepth. Slope can be seen as a driver of soil erosion, but
also as a direct driver of land-use change because
steeper slopes are more difficult to cultivate. Soil
depth can, to a certain extent, be seen as the result of
soil erosion. Soil erosion can affect abandonment in
other ways: it affects soil depth, but it can also affect
abandonment through the loss of nutrients and water-
holding capacity.
The conceptual model presented in Fig. 1
summarises these interrelationships. Slope affects
7/27/2019 Lesvos Bakker Et Al
3/15
M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481 469
Fig. 1. Conceptual model where slope, erosion and soil depth affect abandonment.
abandonment in two ways: (1) via erosion and soil
depth, and (2) directly. Erosion also affects abandon-
ment in two ways: (1) through its effect on soil depth
and (2) through its effect on other processes. In
addition to reducing the soil depth, erosion degrades
soil hydrologic conditions and decreases plant
available water capacity. Sealing of the surface
reduces the infiltration capacity of the soil, thereby
reducing soil moisture. The resulting overland flow
washes away nutrients and organic matter together
with the fine, most fertile soil fraction and the water-
holding capacity will decrease with increasing soil
erosion. Thus, erosion leads to nutrient depletion,
reduction in soil organic carbon and a negative
alteration of the soil physical properties in terms of
nutrient and water-holding capacity. Furthermore,
erosion reduces soil biodiversity and soils becomemore vulnerable to diseases (Lal et al., 1999; Larson
et al., 1985).
This paper attempts to identify the role of soil
erosion as a driver of land-use change in an erosion-
prone area by analysing the relation between soil
erosion and abandonment of cereals in the western part
of Lesvos. The hypothesis will be tested whether the
land-use dynamics were driven by soil erosion. A path
analysis will be carried out in order to evaluate the
validity of the conceptual model presented in Fig. 1
and to assess the relative weights of the various arrowsconnecting slope, erosion, soil depth and the prob-
ability of land-use change.
2. Methodology
2.1. Study area
Lesvos is situated in the north-eastern part of the
Aegean Sea and covers an area of 163,429 ha (Fig. 2).
It is rather hilly with the highest peak 967 m above sea
level. Steep slopes are dominant, covering 63% of the
area. Soils are derived from a variety of parent
materials such as igneous acid rock, ultrabasic rocks,
metamorphic rocks, sedimentary deposits, unconsoli-
dated deposits and recent alluvial deposits, and are
classified as Typic Xerochrept, Lithic Xerochrept
(Eurtric Cambisols) or Lithic Xerothent (Eutric
Regosol) (Kosmas et al., 2000b). In the eastern,
sub-humid region soil depths vary from 0.30 m to
more than 1 m in depth, whereas in the dryer western
part of Lesvos soils are generally shallow (0.15
0.30 m) to very shallow (00.15 m). Here, severely
eroded soils are present with slopes greater than 128,
whereas slightly to moderately eroded soils are found
in the sub-humid zone under the same slope classes
(Kosmas et al., 2000a).The climate is strongly seasonal with spatial
variations in rainfall and high oscillations between
minimum and maximum daily temperatures. Average
annual rainfall is 670 mm, but there is a rainfall
gradient of more than 45% from the east (sub-humid)
to the west (semiarid). The average annual air
temperature is 17.7 8C.
Two land-use periods have been identified (Kosmas
et al., 1999). The first is called the cereals and vines
period and the second one the olives and pastures
period. The mature stage of the second period beganin the mid-20th century, when strong population
migration (a nation-wide phenomenon between the
mid-1950s to the late 1970s) depopulated the island
(Kosmas et al., 1999).
Currently, the dominant land cover of western
Lesvos is shrubland used for grazing (indicated as
rangelands). There is some semi-natural oak forest,
some olive groves and land cultivated with cereals
(mainly wheat (Triticum aestivum L.) and oat (Avena
sativa L.)) (Kosmas et al., 1999).
7/27/2019 Lesvos Bakker Et Al
4/15
470 M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481
Fig. 2. The isle of Lesvos, with areas of interest (AoIs) depicted.
2.2. Methods
2.2.1. Land-use change
As the map of 1996 was made with the purpose of,amongst others, comparison with the 1886 map, both
maps were based on surveys that used the same
classification scheme (although the 1996 map con-
tained a class tobacco, which did not exist in 1886),
and both had the same scale (1:50,000) (Fig. 3). A
difference in accuracy is likely to be the case, as the
survey from 1886 was conducted using topographical
maps, whereas the map from 1996 was made using
topographical maps and aerial photographs. Further-
more, as each map was made by a different person,
differences may exist due to subjective interpretationsof the land-use of the different people. Although these
factors may affect the exact delineation of boundaries
between specific land-uses it is unlikely that they have
had a significant effect on the mapped spatial
distribution of the various land-uses as the difference
between the land-use types that were distinguished are
very clear-cut (Marathianou et al., 2000).
A visual comparison of both maps reveals that
important land-use changes took place between 1886
and 1996. Most of this change occurred during the
1950s: according to the Statistical Service (Ministry of
Agriculture), extensive migration of people to urban
areas occurred during this decade due to, amongst
other things, low land productivity, whilst in the sameperiod the area cultivated with annuals largely
decreased (Kosmas et al., 1999). For the analysis
reported here, both land abandonment (i.e. the
conversion of cereals to rangeland) and land realloca-
tion (the conversion of rangeland to cereals) are
assumed to have taken place in 1956. Clearly this is a
simplification of the actual land-use history of the
area, but the study did not intend to explain the actual
timing of land-use change, rather its spatial distribu-
tion.
From the land-use change map, six smallersubsets were selected that comprised patches of
land that were previously under cereals in 1886, and
for which a part was converted to rangeland
(abandonment) (Fig. 2). Three of these subsets also
included a patch of land that was rangeland in 1886,
of which a part was brought into cereal production in
1996 (reallocation). This allowed samples to be
drawn from the various land-use change trajectories
with similar sizes and which are, therefore, suitable
for statistical tests.
7/27/2019 Lesvos Bakker Et Al
5/15
M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481 471
Fig. 3. Land-use in 1886 and in 1996 (Marathianou et al., 2000; Sifneou, 1996).
2.2.2. Reconstructing slope, erosion and soil depth
Erosion was modelled using the WATEM/SEDEM
model, a spatially distributed model that simulates
erosion and deposition by water and tillage processes
in a two-dimensional landscape (Van Oost et al., 2000;
Van Rompaey et al., 2002). WATEM/SEDEM allows
the integration of landscape structure or the spatial
organisation of different land units and the connec-
tivity between them. In order to avoid problems with
respect to the spatial variability of parameter values
and the inherent uncertainty of these parameter
estimates, WATEM/SEDEM is a relatively simple,
steady-state, topography-driven model. The water
component of WATEM/SEDEM uses an adapted
version of the revised universal soil loss equation
(RUSLE). The tillage component of WATEM/
SEDEM uses a diffusion-type equation for which
the intensity of the tillage process is described by a
7/27/2019 Lesvos Bakker Et Al
6/15
472 M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481
b P
single parameter (the tillage transport coefficient or
ktil-value).
WATEM/SEDEM was initially designed to predict
erosion into the future, but has been modified in this study to calculate erosion into the past. This was done
by calculating forward erosion for 1 year, reversing
the sign of the calculated erosion and deposition
values and modifying the digital terrain model
accordingly. This procedure was then repeated for
the required number of years.
Maps were produced with USLE C- and K-factor
values based on soil and vegetation data. The R-factorwas assumed to be constant for the western part of the
2.2.3. Logistic regression
A dichotomous logistic regression analysis was
carried out to identify the extent to which abandon-
ment and reallocation of cereals has been driven byslope, soil depth and erosion. Dichotomous logistic
regression is a regression technique applicable when
the response variable (in this case, land-use change)
takes only one of two possible values (change or no
change). The logistic model describes the expected
value of the land-use change (i.e. E(change)), or the
probability of land-use change, in terms of the
following logistic formula:
island and was set to 0.041 MJ mm h-1 m-2per year
(Roose, 1996). The P-factor was calculated taking into
account contour ploughing for cereal cultivation for
Echange 1 exp
h
1
k
-
j1
bjXj
\i
the second half of the century (0.6 for slopes of 028,
0.5 for slopes of 268, 0.65 for slopes of 6188 and
0.85 for slopes of 18258). The LS factor wascalculated according to the algorithm of Govers
(1991). Together with a digital terrain model (derivedfrom interpolation of digitised contour lines,Stuckens,
2003), these data layers served as input for the
WATEM/SEDEM model. The tillage transport coeffi-
cient was estimated as 300 kg m-1
for the second half
of the century, and 200 kg m-1
for the first half of the
Prchange (1)
Thus, the probability of land-use change is a function
of k explanatory variables X (in this case slope, erosion
and/or soil depth).
An alternative way to formulate the logistic model
is the logit form. The logit is a transformation of the
probability Pr(change) that is used to make the model
linear and is defined as follows (Kleinbaum et al.,
1998):
century (based on data fromVan Muysen et al., 2000).These values assume that fields were ploughed once a
year with a mouldboard plough, to an average depth of0.20 m during the first half of the century and 0.25 m
logitPrchange] loge
l
Prchangel
Prno change
k
during the second half of the century. During the firsthalf of the century, ploughing speeds were less than
2 km h-1,but exceeded 2 km h-1 in the second half ofthe century.
WATEM/SEDEM was used to reconstruct soil
depths and slope for 1956, the assumed moment of
land-use change, and for 1886, based on the soil
(depth) map for 1996 (Kosmas et al., 2000a).Erosion was calculated for the period 18861956 in
metre of soil loss. For subsequent statistical
analyses, data points were randomly sampled from
these maps (slope and soil depth in 1956, erosion
during the period 18861956) with an average
density of one point per ha. This was done to avoid
dependency between observations of land-use
change. Land-use decisions in the study area are
usually made for individual field parcels which are
smaller than 1 ha.
b0 X
bjXj (2)j1
Using the SAS software (SAS Institute, Inc., 2001),
the models inTable 1 were tested.
2.2.4. Path analysis
In order to validate the conceptual model presented
in Section 1 and to assign a weight to the variousarrows (see Fig. 1), a path analysis was performed.
Path analysis is a techniquemore commonly used in
social sciences that identifies the individual
contribution of each explanatory variable to the model
fit based on correlation coefficients (a measure of
overlap). Traditional path analysis, however, is not
applicable to logistic regression (Davis, 1985).
Instead, likelihood-ratio x2-values are calculated.
Mathematically, likelihood-ratio x2-values have addi-
tive properties, provided that there is no correlation
7/27/2019 Lesvos Bakker Et Al
7/15
M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481 473
slope
erosion
erosion
Table 1
Models tested using logistic regression, which relate slope, erosion
and soil depth to abandonment (a) and reallocation (r),
respectivelyModel Model ID
logit[Pr(abandonment)] = b0 + b1 (slope) 1a
logit[Pr(abandonment)] = b0 + b1 (erosion) 2a
logit[Pr(abandonment)] = b0 + b1 (soil depth) 3a
logit[Pr(abandonment)] = b0 + b1 (slope) 4a
+ b2 (erosion)
logit[Pr(abandonment)] = b0 + b1 (slope) 5a
+ b2 (soil depth)
logit[Pr(abandonment)] = b0 + b1 (erosion) 6a
+ b3 (soil depth)logit[Pr(abandonment)] = b0 + b1 (slope) 7a
When explanatory variables are correlated, the pre-
dictive value of the variables will overlap and the
sum of the x2-values of the explanatory variables will
therefore be larger than the x2
-value of the overallmodel. Therefore, this measure can be used to deter-
mine the extent of overlapping and to evaluate modelsthat attempt to exclude the confounding of variables
(Hamel et al., 1999).
The conceptual model presented inFig. 1 impliesconfounding between slope and erosion and between
erosion and soil depth. If all x2-values were added, the
impact of slope would in part be double-counted: once
+ b2 (erosion) + b3 (soil depth) in the x2-value of slope, and secondly in the x
2-value
logit[Pr(reallocation)] = b0 + b1 (slope) 1r
logit[Pr(reallocation)] = b0 + b1 (erosion) 2r
of erosion. This problem can be solved by decom-posing the explanatory value of slope into a part that
logit[Pr(reallocation)] = b0 + b1 (soil depth) 3r directly affects abandonment (a1x2 ) and a part thatlogit[Pr(reallocation)] = b0 + b1 (slope)
+ b2 (erosion)4r slope
affects abandonment through erosion (a2x2 ).
logit[Pr(reallocation)] = b0 + b1 (slope)
+ b2 (soil depth)
5r Likewise the explanatory value of erosion can bedecomposed into a part that directly affects abandon-
logit[Pr(reallocation)] = b0 + b1 (erosion) 6r ment (a4x2 ) and a part that affects abandonment+ b3 (soil depth)
logit[Pr(reallocation)] = b0 + b1 (slope)through soil depth (a3x2
7r) (Fig. 4). In order to find
+ b2 (erosion) + b3 (soil depth)
between the independent variables (Lemay, 1999).
Where explanatory variables are completely indepen-dent, the overall explanatory value of the modelcomprises the explanatory values of the explanatory
variables, so that the overall model x2
is equal to the
sum of the x2 of the individual independent variables:
x2 var ABC x2 var A x2 var B x2 var C (3)
out how much slope and erosion contribute to each
pathway, the individual and common x2-values of the
applicable combinations of explanatory variables will
be assigned to the various pathways.
It is important to note that the analysis presented
here only allows investigation of the extent to which the data support the proposed hypothesis: as all
relationships are statistical in nature no causality can
be proven. Path analysis merely illuminates which of
two or more competing models, derived from theory,
is most consistent with the pattern of correlations
found in the data (Garson, 2003).
Fig. 4. Conceptual model with x2 indicated.
7/27/2019 Lesvos Bakker Et Al
8/15
474 M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481
Table 2
The physical characteristics of each area of interest (AoI) through time for abandoned land, land still under cereals, reallocated land and land
under permanent rangeland
AoI Change Surface
(ha)
Slope
1956 (8)
Soil loss1886
1956 (m)
Soil depth
1886 (m)
Soil depth
1956 (m)
Soil depth
1996 (m)
1 Abandoned 576 8.0a 0.26 a 0.71a 0.45a 0.26 a
Still cereals 159 2.9 b 0.10 b 1.16 b 1.06 b 0.95 b
Reallocated 126 2.8 a 0.07 a 1.31 a 1.24 a 1.13 a
Permanent rangeland 1362 10.1 b 0.31 b 0.70 b 0.39 b 0.16 b
2 Abandoned 206 6.9 a 0.22 a 1.32 a 1.10 a 1.01 a
Still cereals 65 1.7 b 0.09 b 1.85 b 1.76 b 1.75 b
3 Abandoned 185 9.2 a 0.40 a 1.34 a 0.94 a 0.80 a
Still cereals 46 1.3 b -0.02 b 1.64 b 1.66 b 1.61 b
Reallocated 152 3.0 a 0.08 a 1.68 a 1.60 a 1.56 a
Permanent rangeland 600 8.1 b 0.21 b 0.75 b 0.54 b 0.44 b
4 Abandoned 46 14.6 a 0.41 a 1.60 a 1.19 a 1.12 aStill cereals 26 3.6 b 0.05 b 1.37 b 1.32 b 1.25 b
5 Abandoned 78 2.3 a 0.03 a 0.61 a 0.58 a 0.56 a
Still cereals 386 1.9 b 0.04 a 1.00 b 0.96 b 0.94 b
Reallocated 154 0.9 a 0.01 a 0.96 a 0.95 a 0.93 a
Permanent rangeland 225 2.1 b 0.04 b 0.49 b 0.45 b 0.44 b
6 Abandoned 62 6.9 a 0.16 a 1.33 a 1.17 a 1.12 a
Still cereals 158 1.3 b 0.03 b 1.52 b 1.49 b 1.46 b
AoI: area of interest (see alsoFig. 2). Values not followed by the same character are significantly different at the 0.05 interval.
3. Results
3.1. Erosion, soil depth and slope characteristics per
land-use change trajectory
Table 2 summarises the characteristics of the six
areas of interest (AoIs). For areas that were under
cereals, estimated soil losses in the period 18861956
ranged from 0.05 m for non-erosion-prone land to
0.26 m for erosion-prone land. This is consistent with
the findings of Tsara et al. (2001), who found soil
losses of 0.240.30 m in a 63-year period for a similar
erosion-prone area in Greece.
For all AoIs, soil loss during the period 18861956 is
more severe in areas that were abandoned compared toneighbouring areas that remained under cereals. The
abandoned areas are also located on steeper slopes, and
the soil depths were often very marginal for cereal
cultivation at the assumed moment of change. Likewise,
the areas taken into cultivation suffered less erosion
during the period 18861956 than surrounding areas
that remained under rangeland. Furthermore, slope and
soil depth values were more favourable in the areas
taken into cultivation compared to surrounding areas.
In Table 3, the values of slope, soil loss and
soil depth are shown for all the AoIs. All values
Table 3
The average physical characteristics through time for abandoned land, land still under cereals, reallocated land and land under permanent
rangeland
Change Surface
(ha)
Slope
1956 (8)
Soil loss1886
1956 (m)
Soil depth
1886 (m)
Soil depth
1956 (m)
Soil depth
1996 (m)
Abandoned 1153 7.8 0.26 0.98 0.72 0.55
Still cereals 840 2.0 0.05 1.24 1.19 1.15
Reallocated 432 2.2 0.06 1.32 1.26 1.21
Permanent rangeland 3287 8.6 0.29 0.71 0.42 0.18
7/27/2019 Lesvos Bakker Et Al
9/15
M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481 475
r3
r
770.713***(1)
537.814***(2)959.306***(2)
- . 236.77***(1)
3.871*(1)425.363***(1)
707.447***(1)
474.548***(1)***(2)
0
-232.899***(2)188.593***(1)
04
6
r
795.977***(2)
960.356***(3)
262.034***(2)
426.413***(2)
732.711***(1) 25.264***(1) 2
897.09***(2) 189.643***(2) 4
Table 4
Likelihood-ratio x2-values for all tested models, for abandonment of cereals
Model ID Deviation from
Intersect only 1a 2a 3a 4a 5a 6a1a 698.233***(1) 0
2a 292.912***(1) -405.321***(1) 0
3a 292.194***(1) -406.039***(1) -0.718(1) 0
4a 703.062***(2) 4.829*(1) 410.15***(1) 410.868***(2) 0
5a 803.639***(2) 105.406***(1) 510.727***(2) 511.445***(1) 100.577***(1) 0
6a 493.051***(2) -205.182***(2) 200.139***(1) 200.857***(1) -210.011***(1) -310.588***(1) 0
7a 806.843***(3) 108.61***(2) 513.931***(2) 514.649***(2) 103.781***(1) 3.204(1) 313.792***(1)
Columns showing deviation from intersect only .. . 6acontain the differences in -2 log L values, or likelihood-ratio x2-values between the
various models (Table 1) (including the intersect only). The significance of these differences are tested as x2-values, whereby the degrees of
freedom are determined by the number of additional variables included in the model (when two models were compared that each had one
variable, this was regarded as comparing a one-variable model with an intersect-only model and, therefore, tested as x2 with one degree of
freedom) (Kleinbaum et al., 1998); number between brackets: degrees of freedom. *P < 0.05; **P < 0.01; ***P < 0.001.
between abandoned and still cereals and between
reallocated and permanent rangeland are signifi-
cantly different with a probability of more than
99.999%.
3.2. Logistic regression
Tables 4 and 5 give the goodness of fit of the tested
models (see Table 1) compared to the absence of a
model (intersect only) and to the other models, forabandonment and reallocation, respectively.
For the abandonment of cereals, only model 7a
does not differ significantly from model 5a, meaning
that the variable erosion has no significant added value
to a model already containing soil depth and slope.
The same is true for models 7r and 5r describing the
reallocation of cereals.
The optimal model for the prediction of abandon-
ment is given by
logitPrabandonment] -0:0750:3357
xslope-0:9226
x soil depth (4)
The erosion term is not included since the added pre-
dictive value of this variable is much lower than itsuncertainties: analysis showed that the uncertainty of
the logit is much higher when erosion is included,
whereas the added value as a predictor is quite low.
The added value of slope and soil depth as predictors is
much higher than theuncertaintyinvolved. Furthermore,
the high correlation of erosion with slope causes unreli-
able coefficient estimates (Kleinbaum et al., 1998).
Table 5
Likelihood-ratio x2-values for all tested models, for reallocation of cereals
Model ID Deviation from
Intersect only 1r 2r 3r 4r 5r 6r
1r 533.943***(1) 0
63.266***(1)
Number between brackets: degrees of freedom.*
P < 0.05; **P < 0.01; ***P < 0.001.
21.492***(1) 0
58.163***(1) -163.329***(1) 0
22.542***(1) 1.050(1) 164.379***(1)
7/27/2019 Lesvos Bakker Et Al
10/15
476 M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481
2
x2
x
x2
a3x a4x
a1x a2x x
x
For this model 83.8% of the pairs are concordant. If
it is assumed that the threshold between the
probability of abandonment and the probability of
no abandonment is 0.5 (meaning that when theprobability is higher than 0.5 there will be abandon-
ment, and if lower than 0.5 there will be no
The following x2-values are known from the
logistic analysis (Table 4):
erosion
292:912
soil depth 292:194
abandonment), the model explains 75.6% of theobserved abandonment.
2non-slope-erosion 4:829
The optimal model for the prediction of realloca-
tion is
logitPrreallocation] -2:0695-0:2332
xslope2:0517
x soil depth (5)
For this model, 90.2% of the pairs are concordant and
again, assuming the threshold in probability of,
respectively, reallocation and no reallocation is 0.5,
the model explains 82.0% of the observed realloca-
tion.
x2 -erosion-sd 200:139non
Thus,
a2x2 2 2slope xerosion-xnon-slope-erosion
292:912 - 4:829 288:083
a1x2 2 2
slope xslope-
a2xslope 698:233-
288:083
410:150
a3x2 2 2erosion xsoil depth-xnon-erosion-sd
292:194-200:13992:055
3.3. Pathanalysis
a4x2
n x2
2n292:912-92:055
Following the scheme of Fig. 4 (based on the
erosioerosion-a3xerosio
200:857
conceptual model in Fig. 1), the following set ofequations apply (provided the model excludes con-
founding):
x2 2 2
CompletingEq. (9) with these numbers gives (Fig. 5):
total410:150200:85792:055200:139
903:201
erosion a2xslope xnon-slope-erosion
2erosion
x2 2
2erosion
2
(6)4. Discussion
sd a3xerosion xnon-erosion-sd (7)
2slope
2slope
2slope (8) 4.1. Identifying the effect of erosion on land-use
change
x
2 2 2 2
total a1xslope a4xerosion a3xerosion
2non-erosion-sd (9)
It appears that the abandonment and the realloca-tion of cereals are strongly related to certain physical
properties of the sites. Statistically, the abandonment
The x2-values for non-slope-erosion and non-ero-
sion-soil depth are the added values of, respectively,
erosion and soil depth, compared to a model having,
respectively, only slope and erosion as explanatory
variables, i.e., the difference between the goodness of
fit of the model with only slope and the model with
slope and erosion, and the difference between the
goodness of fit of the model with only erosion and
the model with erosion and soil depth.
and reallocation of cereals are well predicted with
slope and soil depth as explanatory variables. Erosion
has, next to slope and soil depth, no significant added
value in predicting abandonment and reallocation.
The conceptual model (Fig. 1) provides a possible
explanation of why the predictive value of erosion is
not found to be significant. Erosion is in the middle of
a chain of drivers and, therefore, when both slope
and soil depth are included as predictors, erosion itself
7/27/2019 Lesvos Bakker Et Al
11/15
M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481 477
Fig. 5. x2-Value contributions of all explanatory variables.
has no added value as a predictor: the potential effect
of erosion on land-use change is already included in
the variables soil depth and slope.
Slope has significant added value as a predictor
compared to erosion. The path analysis showed that
the total impact of slope was divided approximately
into a third via erosion and soil depth and two-thirds
directly. It appears likely, therefore, that slope affects
abandonment in ways other than through erosion (e.g.
the introduction of machinery that was unsuitable for
steep slopes). Likewise, the effect of erosion on
abandonment can be divided approximately into one-
third via soil depth and two-thirds directly. Erosionaffects, therefore, land-use change not only through
the reduction in soil depth (and therefore rooting depth
and plant available water), but also through the
removal of organic matter and nutrients from the soil,
loss of soil structure and water-holding capacity and
increased stoniness due to selective removal of fine
soil particles (Gonzalez-Hidalgo et al., 1999; Lal et al.,
1999; Larson et al., 1985). Especially in Lesvos, where
the water-holding capacity generally is low (13.1
14.8 mm of water per 10 cm of soil, Kosmas et al.,
2000a) and where stoniness of the soils is problematic
for the workability, it is not surprising that these
direct effects of erosion play a significant role.
It is possible that differences between the maps
could have derived from differences in the survey
methods. This could influence the outcomes of the
regression analysis. Such differences, however, are
more likely to decrease the strength of the statistical
relationship because they would introduce random
noise rather than creating a structured, spatial pattern
that relates to erosion.
4.2. Improving the conceptual model
If the conceptual model in Fig. 1 was a goodrepresentation of reality and accounted for all
confounding between variables, the sum of the x2-
values of the components (903.201) would equal the
x2-value of the total model. However, this is not the
case as the overall x2-value of the total model is
806.043 (Table 4). Thus, not all confounding is
excluded.
It is interesting to explore, therefore, a conceptual
model where slope not only affects soil depth via
erosion, but also directly (seeFig. 6). This could occur
Fig. 6. Revised conceptual model.
7/27/2019 Lesvos Bakker Et Al
12/15
478 M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481
Fig. 7. Revised conceptual model with x2 indicated.
Fig. 8. Revised conceptual model with x2-value contributions of all explanatory variables.
because of: (a) pedogenesis (soil development may be
slower on steep slopes), and/or (b) geomorphology
(long-term erosion rates are related to slope gradient,
leading to a relationship between slope gradient and soil
depth, prior even to the use of land for agriculture). This
long-term erosion effect was not captured by the
WATEM/SEDEM model used here to estimate erosion
rates between 1886 and 1956. Contrary to long-term
erosion rates, erosion rates between 1886 and 1956 are
strongly affected by land-use.
The x2-values associated with the different path-
ways depicted inFig. 6 can then be calculated as above
Fig. 9. Conceptual model for reallocation of cereals with x2-value contributions of all explanatory variables.
7/27/2019 Lesvos Bakker Et Al
13/15
M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481 479
2
(Fig. 7). The sum of the individual model components
is now (Fig. 8):
total 313:792 200:857 92:055
103:781 96:358 806:843
This value is very close to the x2-value of the overall
model (806.04) and, therefore, it can be assumed that
the model excludes most of the significant confound-
ing.
It seems that a significant part of the explanatory
power of the model can be attributed to the erosion/soil depth path, although the improved model does
show a relatively strong, direct linkage between slope
and soil depth. This implies that even if not
Fig. 10. Change-probability maps derived from logistic regression: (a) probability of rangeland being reallocated to cereal cultivation; (b)
probability of abandonment of areas currently under cereals.
7/27/2019 Lesvos Bakker Et Al
14/15
480 M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481
x2
accelerated by human-induced erosion, soil depths
would still be related to a certain extent to slope. The
role of erosion as a driver for land-use change is thus
slightly reduced by this finding.When the same steps are repeated for the
reallocation model (Fig. 9), the contribution of the
separate x2-values is as follows:
total 0:31x533:9430:40x63:266
0:60x63:266422:542960:356
The sum of the components is exactly the same as the
x2-value of the total model, so the model excludesconfounding.
The model explains 82.0% of the observed
reallocation. Although the erosion/soil depth path
contributes significantly to the reallocation model, it
is much less important than in the abandonment
model. Slope itself once more plays a considerable
role. It appears that for the reallocation of cereals,
soil depth has a higher predictive value, relative to
slope, compared to that for the abandonment of
cereals. This implies that in the process of decision-
making, farmers were more concerned with soil
depth than with slope. This is understandable
assuming that decisions were based on farmers
experience with the abandonment of cereals:marginal productivity is more easily associated with
shallow soils than with steep slopes, which are
indirectly related to decreasing productivity.
Furthermore, on shallow soils the effects of erosion
on productivity manifest themselves more strongly
as in shallow soils nutrient and water availability is
lower. Therefore, a further reduction due to erosion
results in a greater relative loss on shallow soils than
on deeper soils.
Furthermore, the path from slope to soil depth is
much stronger for reallocation than for abandonment.Soil depths on rangeland are much less affected by
recent, human-induced soil erosion. Consequently,
erosion is still strongly associated with slope gradient,
due to the effect of slope gradient on soil formation
and on long-term erosion rates.
4.3. Future perspectives
Fig. 10a and b shows the probability of abandon-
ment of areas currently under cereals, and the
probability of rangeland being reallocated to cereals
based on the models derived from the logistic
regression. The probability of abandonment of cereals
is higher than 0.5 almost everywhere, whereas theprobability of reallocation of rangelands to cereals is
almost always lower than 0.5. This implies that under
current conditions almost all fields currently under
cereals will eventually become abandoned since the
current rangelands have no potential for cereal
cultivation. Sustainable cereal cultivation may be
possible in small patches of deep soil that are not
evident because of the resolution of the soil map, but
these patches will be rare.
5. Conclusions
Statistical analysis indicates that the physical
characteristics of the landscape have been important
factors in the abandonment and subsequent realloca-
tion of land under cereals: land with high slope
gradients, high erosion rates and shallow soils tended
to be taken out of cereal production, while new arable
land for cereals is mainly located in areas with deep
soils, low erosion rates and with relatively low slope
gradients.
The strong relationship between abandonment andsoil depth indicates that erosion (i.e. the decrease in
soil depth) may be responsible for abandonment.
Based on the results of the logistic regression, the
goodness of fit of the logistic abandonment model was
attributed as follows: ca. 25% to the direct impact of
erosion, ca. 36% to the erosion/soil depth path and ca.
39% to the direct impact of slope.
For the reallocation, the contribution of the direct
impact of slope to the goodness of fit of the model was
much less (ca. 17%) while the effect of slope through
the erosion/soil depth path was much more important(ca. 80%). This can be explained by the fact that
farmers associated the marginal productivity of the
land they had abandoned previously more to the
shallowness of the soils than to steep slopes.
Consequently, soil depth was a more important
criterion when selecting land for reallocation than
slope gradient.
The results of the analysis presented here are
consistent with the hypothesis that, in the case of west
Lesvos, the spatial pattern of land-use change is
7/27/2019 Lesvos Bakker Et Al
15/15
M.M. Bakker et al. / Agriculture, Ecosystems and Environment 105 (2005) 467481 481
significantly affected by erosion, despite the fact that
logistic regression did not identify modelled erosion
rates as a significant independent variable due to
confounding effects. This illustrates the problem thattraditional statistical analyses may be unable to
identify significant drivers of land-use change.
Furthermore, the results suggest that soil erosion risk
should be included as a variable in more mechanistic
land-use change models (e.g.Rounsevell et al., 2002)
at least when areas with marginal soils are studied.
References
Bakker, M.M., Govers, G., Rounsevell, M.D.A., 2004. The crop
productivityerosion relationship: an analysis based on experi-mental work. Catena 57, 5576.
Davis, J.A., 1985. The logic of causal order, Quantitative Applica-
tions in the Social Sciences Series, 55. Sage Publications,
Richmond, Virginia, pp. 72.
Garson, D., 2003. Path Analysis. NC State University (electronic
source) .
Gonzalez-Hidalgo, J.C., Echeverria, M.T., Vallejo, R.V., 1999.
Selective transport of sediment related to rainfall kinetic energy,
plant cover-stoniness and clearing. Z. Geomorph. NF 43, 255
266.
Govers, G., 1991. Rill erosion on arable land in central Belgium:
rates, controls and predictability. Catena 18, 133155.
Hamel, M.B., Davis, R.B., Teno, J.M., Knaus, W.A., Lynn, J.,
Harrell, F., Galanos, A.N., Wu, A.W., Phillips, R.S., 1999. Older
age, aggressiveness of care, and survival for seriously ill,
hospitalized adults. Ann. Intern. Med. 131, 721728.
Kleinbaum, D.G., Kupper, L.L., Muller, K.E., Nizam, A., 1998.
Applied Regression Analysis and other Multivariable Methods,
Brooks/Cole Publishing Company, .
Kosmas, C., Briassoulis, H., Gerontidis, S., Detsis, B., Marathianou,
M., Dalakou, B., 1999. Medalus III: Lesvos, soilvegetation
interactions. ENV4-CT95-0119. EC Research Project, Brussels.
Kosmas, C., Danalatos, N.G., Gerontidis, S., 2000a. The effect of
land parameters on vegetation performance and degree of ero-
sion under Mediterranean conditions. Catena 40, 317.
Kosmas, C., Gerontidis, S., Marathianou, M., 2000b. The effect of
land use change on soils and vegetation over various lithological
formations on Lesvos (Greece). Catena 40, 5168.
Lal, R., Mokma, D.L., Lowery, B., 1999. Relation between soil
quality and erosion, Soil and Water Conservation Society,Ankeny, IA.
Larson, W.E., Fenton, T.E., Skidmore, E.L., Benbrook, C.M., 1985.
Effects of soil erosion on soil properties as related to crop
productivity and classification. In: Follett, R.F., Steward, B.A.
(Eds.), Soil Erosion and Crop Productivity, ASA-CSSA-SSSA,
Madison, WI, pp. 189211.
Lemay P., 1999. The statistical analysis of dynamics and complexity
in psychology: a configural approach. Master Thesis. Universite
de Lausanne, Lausanne, 1999.
Marathianou, M., Kosmas, C., Gerontidis, S., Detsis, V., 2000. Land-
use evolution and degradation in Lesvos (Greece): a historical
approach. Land Degradation Dev. 11, 6373.
Roose, E., 1996. Land HusbandryComponents and Strategy, Food
and Agriculture Organisation of the United Nations, Rome92-5-
103451-6.
Rounsevell, M.D.A., Annetts, J.E., Audsley, E., Mayr, T., Reginster,
I., 2002. Modelling the spatial distribution of agricultural
land use at the regional scale. Agric. Ecosyst. Environ. 95,
465479.
SAS Institute, Inc., 2001. SAS, Cary, NC, USA.
Sifneou, E., 1996. Lesvos Economic and Social History (1840
1912), Trochalia, Athens.
Stuckens, J., 2003. Contour Gridder .
Tsara, M., Gerontidis, S., Marathianou, M., Kosmas, C., 2001. The
long-term effect of tillage on soil displacement of hilly areas
used for growing wheat in Greece. Soil Use Manage. 17, 113
120.Van Muysen, W., Govers, G., van Oost, K., Van Rompaey, A.J.J.,
2000. The effect of tillage depth, tillage speed and soil con-
dition on chisel tillage erosivity. J. Soil Water Conserv. 55,
355.
Van Oost, K., Govers, G., Desmet, P., 2000. Evaluating the effects of
changes in landscape structure on soil erosion by water and
tillage. Landscape Ecol. 15, 577589.
Van Rompaey, A.J.J., Verstraeten, G., van Oost, K., Govers, G.,
Poesen, J., 2002. Modelling mean annual sediment yield using a
distributed approach. Earth Surf. Process. Landforms 26, 1221
1236.